JP2006023572A - Dialog system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the following problems in the conventional method of presenting a candidate of a recognition result in an easily discriminatable manner to a user with a dialog system; the display is changed according to the score and attribute of the candidate and the difference between both is indefinite depending upon the attribute; also, in the case of a retrieval, the difference is indefinite if the retrieval results are similar, and the user is heretofore required to select the desired retrieval object by recognizing the difference by himself or herself. <P>SOLUTION: The dialog system is equipped with a recognition means 1001 for recognizing the inputted voice, and outputting the candidate list of the recognition result, a difference extraction means 1002 for extracting the difference between the candidates of the recognition result and imparting difference information to the recognition result, a difference definitization response generation means 1003 for generating the response to make the difference definite based on the difference information imparted to the recognition result, and a response presentation means 1004 for presenting the generated response. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technical field in which recognition or search is performed based on user input, and a response is presented to the user by a screen or voice.

  In recent years, an increasing number of electronic devices employ a voice recognition man-machine interface. One example is car navigation. Car navigation specifies a desired point, sets it as a destination, searches for a route from the current position of the vehicle to the destination, displays the route together with a map including the current position, and displays the route to the destination. It has a function to perform guidance.

  The designation of a point by voice operation in car navigation is performed by uttering a unique name of a desired destination, for example, a facility name “Kannai Station”.

  The speech recognition apparatus scores the degree of similarity between the currently set recognition target vocabulary and utterance such as “Kannai Station” as a score, and outputs the recognition word with the highest score as the first candidate. Also, the second and subsequent recognition candidates, such as “Kannai Hall” and “Kannai Station”, are presented to the user simultaneously or sequentially, and the user selects a desired facility by speaking or operating a button from among the recognition result candidates. To do. The details of the spoken dialogue device are omitted because they are described in detail in "Basics of Speech Recognition", directed by L.RABINER, BHJUANG, Satoshi Furui, November 1995, NTT Advanced Technology (hereinafter referred to as Reference 1). To do.

  As a method for changing the display form of a candidate for the user to select a candidate when presenting the candidate, Japanese Patent Laid-Open No. 11-231891 discloses a method for changing the display form according to the similarity (likelihood) of each recognition result. Is disclosed. Japanese Patent Laid-Open No. 2003-167600 and Japanese Patent Laid-Open No. 09-114491 disclose a method of changing the display of the input speech according to the attribute assigned to each result.

  Furthermore, as a situation similar to the above, there are cases where a large number of candidates are generated when a search is performed. For example, for the search term “Sapporo City / Kokurokujo”, “Hokkaido / Sapporo City / Higashi Ward / Kitarokujo East”, “Hokkaido / Sapporo City / Kita Ward / Kitarokujo West”, “Hokkaido / Sapporo City / Kitakujo / “Takuhoku Rokujo” is presented to the user as a candidate. On the other hand, there is a case where a function such as highlighting a search target vocabulary in a text is provided.

Japanese Patent Laid-Open No. 11-231891 JP 2003-167600 A JP 09-114491 A L RABINER, B.H.JUANG, supervised by Sadahiro Furui, “Basics of Speech Recognition”, NTT Advanced Technology, November 1995

  As described above, the conventional method for changing the display form of candidates changes the display according to the scores and attributes of individual candidates when presenting a plurality of candidates. For this reason, depending on the assigned attribute, the difference between the two may not be clear. For example, in the recognition vocabulary of the car navigation system, if there are attributes such as operation commands and facility names, the difference between “Kannai hall” and “Kannai station” having the same attributes is not clear even if they are distinguished and color-coded.

  Also in the case of search, particularly when the search results are similar, highlighting a search keyword does not clarify the difference, and the user needs to recognize the difference himself and select a desired search target. is there.

The dialogue apparatus according to the present invention recognizes input speech and outputs a recognition result candidate list;
A difference extraction means for extracting differences between recognition result candidates and adding difference information to the recognition results;
A difference clarification response generating means for generating a response that makes the difference clear based on the difference information given to the recognition result;
Response presenting means for presenting the generated response.

  According to the present invention, the difference extraction unit adds difference information to the recognition result, the difference clarification response generation unit generates a response that clarifies the difference in the recognition result based on the difference information, and the response presentation unit. Thus, since the response generated by the difference clarification response generation means is presented to the user, the user can easily recognize the difference between candidates.

Hereinafter, an interactive apparatus according to the present invention will be described in detail with reference to the drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram illustrating a configuration of the interactive apparatus according to the first embodiment. In the figure, reference numeral 1001 denotes recognition means, 1002 denotes difference extraction means, 1003 denotes difference clarification response generation means, and 1004 denotes response presentation means. Hereinafter, the operation of each functional block will be described.

  The recognition unit 1001 recognizes the voice input from the microphone, and outputs a recognition result candidate list. The recognition result candidate list is a list of recognition result candidates that the recognition means determines to have a certain validity as a recognition result. Each recognition result consists of a symbol series and a score representing the accuracy of the recognition result. A symbol is a language unit such as a word, a character, or a phoneme, or a recognition unit.

The difference extraction unit 1002 extracts a symbol having a difference between candidates obtained by the recognition unit 1001, and assigns a difference index value, which is difference information, to each candidate symbol.
The difference clarification response generation unit 1003 generates a response to the user that makes the difference clear based on the difference information.
The response presentation unit 1004 presents the response generated by the difference clarification response generation unit 1003 to the user.

  Next, the operation flow of the first embodiment will be described with reference to the flowchart of FIG. Here, the object of recognition is the name of the facility, and the difference between the two is clarified, taking as an example the case where two recognition candidates “XX department store / Yokohama / store” and “XX department store / Yokohama / store” are obtained. The procedure for returning a response that has been converted is described.

First, the response presenting means 1004 confirms that there is no input from the user in the recognizing means 1001, and prompts the user to input by the guidance such as “Please speak and input the destination” (step S11). ).
When the user inputs the destination by speaking, the recognition unit 1001 accepts the user's input and performs analysis / recognition processing (step S12). Specifically, the input speech is sampled at a time period T, and a signal sequence x [t] for the discretized time t is obtained. This is converted into a feature vector (for example, 12-dimensional mel cepstrum) in units of 10 ms. Further, the obtained feature vector is collated with an acoustic dictionary and a language dictionary (not shown) provided in the dialogue apparatus, and n recognition result candidates r [1] to r [n] having a high likelihood of collation are obtained. obtain. Each recognition result candidate is a string of symbols that is a recognition unit. For each recognition result candidate, scores s [1] to s [n] representing validity are obtained.

  Here, as a recognition result, two candidates r [1] = “XX department store / Yokohama / store” and r [2] = “XX department store / Yokohama / store” are obtained and input to the difference extraction means 1002. The However, the middle point “·” is a symbol delimiter.

Next, the difference extraction unit 1002 extracts symbols that differ between the obtained candidates, and assigns a difference index value to each candidate symbol (step S13). As the difference index value, for example, d (c; r [1] .. r [n]) = nk is assigned to a symbol c that appears in k candidates among n candidates. In this case, a symbol (common element) that appears in all candidates is assigned 0, and a symbol that appears only in one specific candidate is assigned n-1. In the above example, n = 2, the symbols “Yokohama” and “Store” appear as two candidates (k = 2), and “XX department store” and “XX department store” appear as one candidate (k = 1). Therefore, it becomes as follows.
d (XX Department Store; r [1], r [2]) = 2-1 = 1
d (○ × Department store; r [1], r [2]) = 2-1 = 1
d (Yokohama store; r [1], r [2]) = 2-2 = 0

  Appropriate post-processing such as determining the appropriate threshold value and discretizing it into 0 or 1 may be added to the obtained numerical value.

  Next, the difference clarification response generation unit 1003 reads the recognition result candidate and the difference index value given to the symbols constituting the candidate, and generates a response that clarifies the difference in the recognition result based on the difference index value. (Step S14). In the above-described example, the difference index value of the symbol is referred to and clarified by changing the output form of “XX department store” and “XX department store” having a difference index value of 1. In the case of screen output using characters, emphasis is given by changing the font or adding modifications such as parentheses. For example, changing the font means emphasizing the target character string by changing the color, background color, underline, italic, bold, font, and the like.

  As a result, it is possible to respond by clarifying the difference by enclosing it in parentheses “·”, such as “recognition result: (1)“ XX department store ”Yokohama store, (2)“ XX department store ”Yokohama store”. Further, as a result of the difference extraction, the difference can be clarified by changing the expression and display position of the same part. In the reference example, “recognition result: (1) XX department store Yokohama store, (2) XX department store 〃” “Yokohama store” with a difference index value of 0 indicates that it is the same after the second time. It may be replaced with a symbol or the like. In addition, the difference can be further clarified by displaying “〃” in correspondence with “Yokohama store” and displaying the same positions.

  In addition, in the case of a response by speech generated by speech synthesis or the like, this method indicates on the text to be input to the speech synthesizer, prosodic information such as speech height, loudness, utterance speed, pose provision, etc. Clarify the differences. For example, in the above-mentioned example, it is possible to emphasize by posing once in front of “○ × Department Store” and speaking slowly while raising the voice. In the case of text-to-speech synthesis, markup such as pitch increase is added to the text of the response speech in order to output the speech. The addition of the difference may be binary or may be multi-valued or continuous depending on the magnitude of the difference.

  Finally, the response presentation unit 1004 presents the response generated by the difference clarification response generation unit 1003 to the user (step S15).

  As described above, according to the present embodiment, the difference index value is assigned to the recognition result candidate and the symbols constituting the candidate by the difference extraction means 1002, and the difference clarification response generation means 1003 is assigned. Is generated, a response that clarifies the difference in the recognition result is generated based on the difference index value, and the response generated by the difference clarified response generating unit 1003 is presented to the user by the response presenting unit 1004. Can recognize differences between candidates.

Embodiment 2. FIG.
FIG. 3 is a block diagram showing a configuration of the interactive apparatus according to the second embodiment. In the figure, 2001 is a name database, 2002 is a search means for searching the name database 2001, 1002 is a difference extraction means, 1003 is a difference clarification response generation means, and 1004 is a response presentation means. Hereinafter, the operation of each functional block will be described. However, what has already been described is assigned the same number and description thereof is omitted.

  The name database 2001 is a database for extracting information associated with an input keyword. FIG. 4 is an example of a search keyword and a search result in the database regarding the facility name. The search keyword is a part of the official name of the facility name or a common expression expressed in other words, and the official name is obtained as a search result. The official name is stored as a sequence of symbols separated by a midpoint “·”.

  The search means 2002 searches the name database 2001 using the input search formula or search keyword, and outputs a list of search results. The search may be a partial match with the official name or a reference to a previously created table. The search expression may be a combination of a plurality of search conditions.

  Next, the operation flow of the present embodiment will be described with reference to the flowchart of FIG. In the present embodiment, for the database search example shown in FIG. 4, a procedure for returning a response that clarifies the difference in the search result list with reference to the name database 2002 will be described.

  First, the response presentation means 1004 performs a process of prompting the user to input based on a guidance such as “Please enter search conditions” (step S21).

  Next, the search means 2002 receives the search command r [0] = “XX department store” input by the user in accordance with step S21, searches the name database 2001, and obtains a list of search results (step S22). .

  For example, in order to check the official name of the facility, the database 2001 that can obtain the official name from the common name of the facility is accessed, and the formal name of the possible facility is returned. In the example of FIG. 4, for the keyword “XX department store”, the official names r [1] = “XX department store / Yokohama store” and r [2] = “Kawasaki / XX department store” are obtained.

Next, the difference extraction means 1002 extracts symbols that differ between the obtained candidates, and assigns a difference index value to each candidate symbol (step S23). For example, as a difference index value, d (c; r [1] .. r [n]) = nk is assigned to a symbol c that appears in k candidates among n candidates. In this case, it becomes as follows.
d (XX Department Store; r [1], r [2]) = 2-2 = 0
d (Yokohama store; r [1], r [2]) = 2-1 = 1
d (Kawasaki; r [1], r [2]) = 2-1 = 1

  Next, the difference clarification response generation unit 1003 reads the candidate search results and the difference index values given to the symbols constituting the candidates, and sends a response that clarifies the difference between the search results based on the difference index values. Generate (step S24). As a result, for example, the difference can be clarified by enclosing it in parentheses “·” like “search result: (1) XX department store“ Yokohama store ”, (2)“ Kawasaki ”XX department store”.

  Finally, the response presentation unit 1004 presents the response generated by the difference clarification response generation unit to the user (step S25).

  As a result, the difference in search results is clarified in the generated response. For this reason, the user can easily select a desired candidate.

Although the difference between the search results has been clarified in the above example, a search keyword may be given as an input to the difference extraction unit 1002.
In this case, r [0] = “XX department store” is input to the difference extraction unit 1002, and the difference index value given to the symbol by the above-described difference index value assigning method is as follows. As a result, other than the input search keyword r [0] = “XX department store” becomes clearer.
d (XX department store; r [0, r [1], r [2]) = 3-3 = 0
d (Yokohama store; r [0], r [1], r [2]) = 3-1 = 2
d (Kawasaki; r [0], r [1], r [1]) = 3-1 = 2

  The database search result may be composed of a plurality of semantically separated fields. In this case, a response can be generated by obtaining a symbol difference for each field. If priority is given to a field and there is a difference in a field with a high priority, the difference contents of a field with a low priority may not be presented.

Embodiment 3 FIG.
FIG. 6 is a block diagram showing a configuration of the interactive apparatus according to the third embodiment. In the figure, 1001 is a recognition means, 2001 is a name database, 2002 is a search means, 1002 is a difference extraction means, 1003 is a difference clarification response generation means, and 1004 is a response presentation means. Each functional block has already been described. Next, the operation will be described with reference to the flowchart shown in FIG. 7, taking as an example the case where the utterance “XX department store” is made.

  First, the response presenting means 1004 sends a message prompting the user to input using the text “Please speak” or voice guidance (step S31).

Next, the recognizing unit 1001 receives the input uttered by the user, and analyzes and recognizes the input voice (step S32). Specifically, the input speech is sampled with a time period T to obtain a signal sequence x [t] for time t. This is converted into a feature vector (for example, 12-dimensional mel cepstrum) in units of 10 ms. Further, the obtained feature vector is collated with an acoustic dictionary and a language dictionary (not shown) provided in the dialogue apparatus, and n recognition result candidates r [1] to r [n] having a high likelihood of collation are obtained. obtain.
Each recognition result candidate is a string of symbols that is a recognition unit. For each recognition result candidate, scores s [1] to s [n] representing validity are obtained. Here, it is assumed that two recognition results “XX department store” and “XX department store” are obtained.

  Next, the search means 2002 uses the recognition result candidate output obtained by the recognition means 1001 as an input and searches the name database 2001 to obtain a list of search results (step S33). The search is performed for the number of recognition result candidates. For example, in order to check the correct facility name, the name database 2001 is accessed, and a list of registered names of facilities that are partially matched with the keyword input by the user is returned. As a result, as shown in FIG. 4, for example, if the recognition result candidate “XX department store” is a search keyword, r [1] = “XX department store / Yokohama / store”, r [2] = “Kawasaki / Obtained the official name “XX Department Store”. Also, if the recognition result candidate “XX department store” is a search keyword, the official name r [3] = “XX department store / Yokohama / store” is obtained.

Next, the difference extraction unit 1002 extracts a symbol having a difference between the obtained candidates, and assigns a difference index value to each candidate symbol (step S34). When d (c; r [1] .. r [n]) = nk is assigned to a symbol c that appears in k candidates out of n candidates, the difference index value is as follows.
d (XX Department Store; r [1], r [2], r [3]) = 3-2 = 1
d (○ × Department store; r [1], r [2], r [3]) = 3-1 = 2
d (Yokohama store; r [1], r [2], r [3]) = 3-2 = 1
d (Kawasaki; r [1], r [2], r [3]) = 3-1 = 2

  Next, the difference clarification response generation unit 1003 reads the recognition result candidate and the difference index value given to the symbols constituting the candidate, and generates a response that clarifies the difference in the search result based on the difference index value. (Step S35). As a result, for example, search results: (1) XX department store “Yokohama store”, (2) “Kawasaki” XX department store, and (3) “XX department store” Yokohama store ”are displayed with clarified differences. Generate a response. In the case of generating a response by voice, a response in which the difference is clarified can be generated by the same procedure as in the first embodiment.

  Finally, the response presentation unit 1004 presents the response generated by the difference clarification response generation unit 1003 to the user (step S36).

Embodiment 4 FIG.
FIG. 8 is a block diagram showing a configuration of the interactive apparatus according to the fourth embodiment. In the figure, 1001 is a recognition means, 2001 is a name database, 2002 is a search means, 4001 is a candidate-to-difference extraction means as a difference extraction means, 4002 is a candidate rearrangement means, 1003 is a difference clarification response generation means, and 1004 is Represents response presentation means. In the present embodiment, by rearranging so that the difference between two adjacent candidates is reduced and presenting the response, a response that makes the difference clear can be returned to the user. Hereinafter, each functional block will be described. However, the functional blocks that have already been described are assigned the same numbers, and descriptions thereof are omitted.

  The candidate pair difference extraction unit 4001 reads a list of candidates composed of one or more symbol series, and when there are a plurality of candidates, obtains and outputs a difference distance and a difference content indicating a degree of difference between two appropriate candidates.

  Candidate rearranging means 4002 reads the candidate list, the difference distance between candidates, and the content of the difference, and rearranges the order of the candidate list so as to minimize the sum of the difference distances for adjacent candidate pairs under the constraint conditions.

  Next, the operation flow of the present embodiment will be described with reference to the flowchart of FIG. Here, it is assumed that recognition result candidates “XX department store” and “XX department store” are obtained as outputs of the recognition unit 1001 as shown in FIG.

  First, the response presenting means 1004 sends a message prompting the user to input using the text “Please speak” or voice guidance (step S41).

  Next, the recognizing unit 1001 receives the input uttered by the user, and analyzes and recognizes the input voice (step S42). Here, it is assumed that two recognition results “XX department store” and “XX department store” are obtained for the input voice.

  Next, the search means 2002 uses the recognition result candidate obtained from the recognition means 1001 as an input and searches the name database 2001 to obtain a list of search results (step S43). The search by the search means 2002 is performed for the number of recognition result candidates. For example, in order to find out the correct facility name, the name database 2001 is accessed, and a list of registered names of possible facilities that partially match the keyword entered by the user is returned. As a result, as shown in FIG. 4, when the recognition result candidate “XX department store” is a search keyword, r [1] = “XX department store / Yokohama store”, r [2] = “Kawasaki / XX department store” To get the official name. Also, if the recognition result candidate “XX department store” is a search keyword, the official name r [3] = “XX department store / Yokohama shop” is obtained.

Next, the candidate pair difference extracting means 4001 obtains a suitable difference distance and difference contents between two candidates (step S44). For example, the difference distance and the difference contents are calculated by assigning d (c; r [1], r [2]) = 1 to the symbol c that appears only in one of the two candidates, and it appears in both Assign d (c; r [1], r [2]) = 0 to the symbol c, and set the number of symbols for which d (c; r [1], r [2]) = 1 as the difference distance, the symbol Can be the difference. When r [1] = “XX department store / Yokohama store” and r [2] = “○ × department store / Yokohama store”, it is as follows.
d (XX Department Store; r [1], r [2]) = 2-1 = 1
d (○ × Department store; r [1], r [2]) = 2-1 = 1
d (Yokohama store; r [1], r [2]) = 2-2 = 0
Therefore, the difference distance D (r [1], r [2]) = 2, the difference content {XX department store, XX department store}. Similarly, as shown in FIG. 10, the difference distance and difference contents of the candidate pair are obtained.

  There are n (n-1) candidate pairs in consideration of the order of n candidates. In the example of FIG. 10, since D (X, Y) = D (Y, X), half can be reduced. In addition, in the candidate rearranging means 4002, for example, when there is an order restriction such that the search result of the first recognition result is presented first, it is only necessary to calculate a necessary candidate pair. The order of recognition result candidates for different recognition results may not be changed. Furthermore, it is not necessary to rearrange at all. In this case, the difference distance does not need to be calculated, and only the difference between adjacent candidates is clarified.

  Next, the candidate rearranging means 4002 rearranges the candidate pairs based on the obtained difference distances of the candidate pairs so that the sum of the difference distances of the adjacent candidate pairs after the rearrangement is minimized (step S45). . In the example of FIG. 10, when “r [2] · r [1] · r [3]” are arranged, the sum of the difference distances between adjacent candidate pairs is D (r [2], r [1]). + D (r [1], r [3]) = 4. In this example, the difference distance is the smallest in the above example and in the reverse order. If the distance is the same, you can choose either one. In addition, if the presentation order is limited and r [1] or r [2] of the first recognition result is displayed first, the former is selected.

  Next, the difference clarification response generation means 1003 reads the recognition result candidate and the difference index value given to the symbols constituting the candidate, and generates a response that clarifies the difference in the search results based on the difference contents. (Step S46). As a result, for example, the search results: (1) “Kawasaki” XX department store, (2) XX department store “Yokohama store”, (3) “XX department store” Yokohama store ” Produce responses that are ordered to minimize differences between adjacent candidates. In the case of generating a response by voice, a response in which the difference is clarified can be generated by the same procedure as in the first embodiment.

  Finally, the response presentation unit 1004 presents the response generated by the difference clarification response generation unit to the user (step S47).

  As a result, it is possible to show the user a response that clarifies the difference in the search results based on the plurality of candidates obtained as recognition candidates. For this reason, the user can easily select a desired candidate.

Embodiment 5. FIG.
FIG. 11 is a block diagram illustrating a configuration of the interactive apparatus according to the fifth embodiment. In the figure, 1001 is a recognition means, 2001 is a hierarchical structure database, 2002 is a search means, 5001 is a candidate versus hierarchical structure difference extraction means as a difference extraction means, 4002 is a candidate rearranging means, and 1003 is a difference clarification response generation means , 1004 represents response presentation means. In the present embodiment, when the search result of the database has a hierarchical structure, rearranging so that the difference between two adjacent candidates is reduced and presenting the response, a response that makes the difference clear to the user Can return. Hereinafter, each functional block will be described. However, the functional blocks that have already been described are assigned the same numbers, and descriptions thereof are omitted.

  Candidate-to-hierarchical structure difference extraction means 5001 reads a candidate list consisting of one or more symbol sequences, and if there are multiple candidates, a difference distance and difference representing the degree of difference assuming a hierarchical structure between two appropriate candidates Find and output the contents.

  Next, the operation flow of this embodiment will be described with reference to the flowchart of FIG. Here, as shown in FIG. 13, it is assumed that address information, which is an example of a hierarchical structure, is obtained as a result of searching the hierarchical structure database 2001 using the output of the recognition means 1001 as the input of the searching means 2002.

  First, the response presenting means 1004 sends a message prompting the user to input using the text “Please speak” or voice guidance (step S51).

  Next, the recognizing unit 1001 receives the input uttered by the user, and analyzes and recognizes it (step S52). Here, it is assumed that two recognition results “XX department store” and “XX department store” are obtained.

  Next, the candidate-to-hierarchical structure difference extracting unit 5001 obtains a suitable difference distance and difference content between two candidates (step S54). In the difference extraction, the search result is assumed to have a hierarchical structure as shown in FIG. For this reason, the hierarchy after the symbol in which the difference in the hierarchical structure occurs is not effective for clarifying the difference. Therefore, the difference distance between the two addresses is, for example, (maximum number of hierarchies L) − (hierarchy number where the difference occurs) +1. Further, the content of the symbol of the hierarchy number where the difference has occurred is used as the difference content. For example, when the maximum number of hierarchies is L = 3, the difference distance between r [1] = “Kanagawa Prefecture / Yokohama City / Nishi Ward” and r [2] = “Kanagawa Prefecture / Kawasaki City / Kawasaki Ward” is 2. (Yokohama City, Kawasaki City).

  There are n (n-1) candidate pairs in consideration of the order of n candidates. In the example of FIG. 10, since D (X, Y) = D (Y, X), half can be reduced. Further, if there is an order restriction such that the candidate rearranging means 4002 presents the search result of the first recognition result first, it is sufficient to calculate only the necessary candidate pairs. The order of recognition result candidates for different recognition results may not be changed. Furthermore, it is not necessary to rearrange at all. In this case, the difference distance does not need to be calculated, and only the difference between adjacent candidates is clarified.

  Next, the candidate rearranging means 4002 rearranges the candidate pairs based on the obtained difference distances of the candidate pairs so that the sum of the difference distances of the adjacent candidate pairs after the rearrangement is minimized (step S55). . Here, the addresses of different recognition results are not rearranged, and are applied only to r [1] to r [3]. In this example, it is sufficient that r [1] and r [3] are adjacent to each other. For example, the order is r [3] · r [1] · r [2].

  Next, the difference clarification response generation unit 1003 reads the recognition result candidate and the difference index value given to the symbols constituting the candidate, and generates a response that clarifies the difference in the search result based on the difference index value. (Step S46). As a result, for example, "Search results: XX department store (1) Kanagawa Prefecture / Yokohama City / Totsuka Ward, (2) Nishi Ward, (3) Kawasaki City / Kawasaki Ward, XX Department Store (4) Kanagawa Prefecture / Yokohama City / A response is generated in which partial differences are clarified, such as “Nishi-ku”, and are arranged so that differences between adjacent candidates are minimized. The same procedure can be used to clarify the difference when generating a response by voice. For example, “There are three XX department stores. The addresses are Kanagawa Prefecture, Yokohama City, Totsuka Ward and Nishi Ward, Kawasaki City, Kawasaki Ward. There is one XX Department Store. The address is Kanagawa Prefecture, Yokohama City. A response such as “Nishi Ward” can be generated.

  Finally, the response presentation unit 1004 presents the response generated by the difference clarification response generation unit 1003 to the user (step S57).

  As a result, it is possible to show the user a response that clarifies the difference in the hierarchical database search results based on the plurality of candidates obtained as recognition candidates. For this reason, the user can easily select a desired candidate.

  In the first to fifth embodiments, the database 2001 is not provided in the dialogue apparatus, but may be an external database connected by a communication line or the like.

  The difference extraction means extracts the difference between the recognition result candidates and gives difference information to the recognition result, and the difference clarification response generation means generates a response that makes the difference clear based on the difference information given to the recognition result. Therefore, the user can easily distinguish the difference between the recognition result candidates, and the operability of the user is improved by applying it to car navigation or the like.

It is a block diagram which shows the structure of the dialogue apparatus which concerns on Embodiment 1 of this invention. 3 is a flowchart of an operation flow according to the first embodiment. It is a block diagram which shows the structure of the dialogue apparatus which concerns on Embodiment 2 of this invention. It is explanatory drawing which shows the example of the search keyword of the database regarding a facility name, and a search result. 6 is a flowchart for explaining an operation flow of the second embodiment. It is a block diagram which shows the structure of the dialogue apparatus which concerns on Embodiment 3 of this invention. 12 is a flowchart for explaining an operation flow of the third embodiment. It is a block diagram which shows the structure of the dialogue apparatus which concerns on Embodiment 4 of this invention. 10 is a flowchart for explaining an operation flow of the fourth embodiment. It is explanatory drawing of the difference distance and difference content of a candidate pair. It is a block diagram which shows the structure of the dialogue apparatus which concerns on Embodiment 5 of this invention. 10 is a flowchart for explaining an operation flow of the fifth embodiment. FIG. 10 is an explanatory diagram of a recognition result by a recognition unit in the fifth embodiment. FIG. 16 is an explanatory diagram of search results in the fifth embodiment.

Explanation of symbols

  1001: Recognition means, 1002: Difference extraction means, 1003: Difference clarification response generation means, 1004: Response presentation means, 2001: Name database, 2002: Search means, 4001: Candidate versus difference extraction means, 4002: Candidate rearrangement means 5001: Candidate versus hierarchical structure difference extraction means.

Claims (11)

Recognizing means for recognizing input speech and outputting a candidate list of recognition results;
A difference extraction means for extracting differences between recognition result candidates and adding difference information to the recognition results;
A difference clarification response generating means for generating a response that makes the difference clear based on the difference information given to the recognition result;
A dialogue apparatus comprising response presentation means for presenting a generated response. Search means for searching the database using the input search keyword and outputting a plurality of search results;
A difference extracting means for extracting the difference between the plurality of obtained search results and the difference between the search result and the input search keyword and adding difference information to the search result;
A difference clarification response generating means for generating a response that makes the difference clear based on the difference information given to the search result;
A dialogue apparatus comprising response presentation means for presenting a generated response. Recognizing means for recognizing input speech and outputting a candidate list of recognition results;
3. The interactive apparatus according to claim 2, wherein the search means uses the recognition result candidate list as an input search keyword. The difference extraction means is configured to extract a difference between search results and add a difference distance and a difference content to the recognition result.
Sorting means for rearranging the order of the candidates to be presented so that the difference is clear based on the difference distance given by the difference extracting means,
4. The dialogue apparatus according to claim 3, wherein the difference clarification response generation unit is configured to generate a response so that a difference between adjacent candidates is clear.   5. The interactive apparatus according to claim 4, wherein the search means is configured to search the database and output a search result having a hierarchical structure using the recognition result candidate list as an input search keyword.   The difference extraction means assigns a difference index value to the symbols constituting the candidate, and the difference clarification response generation means generates a response for changing the presentation form based on the assigned difference index value. 6. The interactive apparatus according to any one of 5.   The difference clarification response generating means generates a response that emphasizes the difference between candidates by changing at least one of the display position and the display form for each symbol constituting the candidate according to the degree of difference in generating the response expression by text / screen. The interactive device according to claim 1, wherein the interactive device is generated.   Difference clarification response generation means generates a response that emphasizes the difference part by replacing the common symbols that make up the candidate in the generation of the response expression by text / screen with the same symbol or blank after the second appearance An interactive apparatus according to claim 1, wherein   The difference clarification response generation means generates at least one of a voice pitch, a loudness level, a speech rate, and a pause for each symbol constituting a candidate according to the degree of difference in generating a response expression by speech. 6. The dialogue apparatus according to claim 1, wherein the difference is emphasized by changing.   The search means outputs a search result including at least one of an official name of the facility name, a facility genre, and an address when the input search keyword is a common name of the facility name. The interactive device according to the above.   The search means classifies the search results of the database into a plurality of fields that are classified from a predetermined point of view, and the response presenting means determines if there is a difference in the search results of the fields with high priority. 6. The interactive apparatus according to claim 2, wherein a difference in a low priority field is not presented.

Images